[llvm] [AMDGPU] PromoteAlloca: handle out-of-bounds GEP for shufflevector (PR #139700)

[Matt Arsenault via llvm-commits]

================
@@ -0,0 +1,45 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt -mtriple amdgcn -passes=amdgpu-promote-alloca-to-vector -S < %s | FileCheck %s
+
+; Skip promote-alloca in case of an index which is known to be out of bounds.
+
+define amdgpu_kernel void @out_of_bounds() {
+; CHECK-LABEL: define amdgpu_kernel void @out_of_bounds() {
+; CHECK-NEXT:    [[PTR:%.*]] = freeze <4 x float> poison
+; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <4 x float> [[PTR]], <4 x float> poison, <4 x i32> <i32 poison, i32 poison, i32 2, i32 3>
+; CHECK-NEXT:    ret void
+;
+  %ptr = alloca [4 x float], align 4, addrspace(5)
+  %elem_ptr = getelementptr [4 x float], ptr addrspace(5) %ptr, i32 0, i32 42
+  call void @llvm.memcpy.p5.p5.i32(ptr addrspace(5) %ptr, ptr addrspace(5) %elem_ptr, i32 8, i1 false)
+  ret void
+}
+
+define amdgpu_kernel void @memcpy_partially_out_of_bounds() {
+; CHECK-LABEL: define amdgpu_kernel void @partially_out_of_bounds() {
+; CHECK-NEXT:    [[PTR:%.*]] = freeze <3 x float> poison
+; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <3 x float> [[PTR]], <3 x float> poison, <3 x i32> <i32 2, i32 poison, i32 2>
+; CHECK-NEXT:    ret void
+;
+  %ptr = alloca [3 x float], align 4, addrspace(5)
+  %elem_ptr = getelementptr [3 x float], ptr addrspace(5) %ptr, i32 0, i32 2
+  call void @llvm.memcpy.p5.p5.i32(ptr addrspace(5) %ptr, ptr addrspace(5) %elem_ptr, i32 8, i1 false)
+  ret void
+}
+
+define amdgpu_kernel void @in_bounds() {
+; CHECK-LABEL: define amdgpu_kernel void @in_bounds() {
+; CHECK-NEXT:    [[PTR:%.*]] = freeze <4 x float> poison
+; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <4 x float> [[PTR]], <4 x float> poison, <4 x i32> <i32 2, i32 3, i32 2, i32 3>
+; CHECK-NEXT:    ret void
+;
+  %ptr = alloca [4 x float], align 4, addrspace(5)
+  %elem_ptr = getelementptr [4 x float], ptr addrspace(5) %ptr, i32 0, i32 2
+  call void @llvm.memcpy.p5.p5.i32(ptr addrspace(5) %ptr, ptr addrspace(5) %elem_ptr, i32 8, i1 false)
+  ret void
+}
+
+; Function Attrs: nocallback nofree nounwind willreturn memory(argmem: readwrite)
----------------
arsenm wrote:

```suggestion
```

https://github.com/llvm/llvm-project/pull/139700